In actual production line operation, we have found that AOI inspection must be performed immediately after Reflow Soldering cooling down to below 40°C; otherwise, residual heat will cause slight deformation of the PCB surface, which in turn leads to positioning offset. The JiajinYuan AOI inspection system can automatically correct thermal expansion errors within the range of ±0.08mm through a built-in temperature compensation algorithm, which is more reliable than simply relying on mechanical fixtures for repeated positioning. When encountering PCBs with OSP surface treatment, the sensitivity of AOI inspection to pad reflection will significantly increase. At this time, the incident angle of the ring light source must be synchronously adjusted to 32 degrees and the polarization filter module must be enabled; otherwise, the grayscale difference between cold solder joints and normal solder joints will be compressed to less than 18 grayscale units.

The recognition sensitivity of JiajinYuan AOI inspection for tombstone defects of 0402 components depends on the coordinated matching of side fill light intensity and image sampling frame rate. If the conveyor belt speed at the reflow oven outlet is set to 78cm/min and the AOI inspection frame rate is not synchronously increased to 32fps, the PCB displacement during single-frame exposure will exceed 0.035mm, causing edge blur. At this time, the system will misjudge real tombstones as slight offsets. We once used the same verification board for continuous testing under different temperature and humidity environments. When the workshop humidity suddenly rose from 45%RH to 68%RH, the capture rate of AOI inspection for QFN device pin cold solder joints dropped from 99.1% to 92.7%. The root cause is that water vapor forms a submicron-scale condensation film on the optical lens surface, changing the refraction path. Therefore, the JiajinYuan AOI inspection equipment strictly requires that the humidity in the lens compartment be maintained at 30±5%RH and wiped once per shift with a dust-free cloth dipped in isopropanol.

For invisible areas such as BGA bottom solder joints, although AOI inspection cannot penetrate the package for imaging, it can indirectly infer the risk of cold solder joints through the triple feature coupling analysis of the Solder Paste collapse shape around the solder ball, the distribution of surrounding Flux residue, and the thermal diffusion shadow of adjacent pads. For example, when the flux carbonization trace at the corresponding position of a certain BGA solder ball presents an asymmetric fan-shaped expansion, the infrared thermal image temperature difference between two adjacent pads exceeds 3.6°C, and the curvature radius of the solder paste edge contour is less than 27μm, the JiajinYuan AOI inspection system will trigger a secondary re-inspection process and mark 'suspected cold solder joint' for manual confirmation. This logic based on weighted judgment of multi-dimensional image features is much closer to the real production line defect distribution law than traditional AOI inspection that simply relies on single threshold segmentation.

When the JiajinYuan AOI inspection equipment is used for mixed production of hot air leveling boards and immersion gold boards, the corresponding reflectivity calibration template must be loaded in advance, because the reflectivity of the immersion gold layer to the blue light band (450nm) is 41% lower than that of the hot air leveling layer. If the template is not switched and the hot air leveling parameters are directly used, it will lead to broken solder joint contour extraction, especially at 0.3mm pitch QFP pins, where three consecutive pins are easily misjudged as open circuits. We once observed on a dual-track SMT line that when the left track uses immersion gold boards and the right track uses hot air leveling boards, if the AOI inspection does not bind the track ID to automatically call the corresponding template, its overall false alarm rate will rise to 4.9%. However, as long as the template switching is completed within 0.8 seconds after Mark point recognition, the false alarm can be suppressed to within 1.3%.